Last updated: 2026-06-27

Changelog — June 27, 2026

Claude Code shipped v2.1.195, a correctness-and-reliability release. The one item to act on: hook matchers with hyphenated identifiers (like code-reviewer or mcp__brave-search) were accidentally substring-matching and now exact-match — check your hook configs, and use mcp__brave-search__.* to match every tool from a hyphenated MCP server. It also adds CLAUDE_CODE_DISABLE_MOUSE_CLICKS, fixes voice dictation, tightens external-plugin install consent, and clears a batch of background-agent bugs. NemoClaw landed user-facing changes on main (no new tag): Deep Agents CLI aliases, a destroy that finally wipes per-sandbox persistent state, host-authoritative shields status, and an Ollama non-interactive fix. (No new releases from Kilo Code, IronClaw, or Hermes today; the Claude apps/API release-note pages updated but published no substantive new items beyond yesterday's.)

2026-06-27 Claude Code v2.1.195

A bug-fix release weighted toward hook matching, voice dictation, plugins, and background agents. One change can quietly alter behavior on upgrade — read it first:

  • Hyphenated hook matchers now exact-match — check your hooks: matchers with hyphenated identifiers (e.g. code-reviewer, mcp__brave-search) were accidentally substring-matching, so a matcher could fire on more tools than intended. They now exact-match. If you relied on the loose behavior to catch every tool from a hyphenated MCP server, switch to an explicit wildcard like mcp__brave-search__.*.
  • New CLAUDE_CODE_DISABLE_MOUSE_CLICKS setting: disables mouse click/drag/hover in fullscreen mode while keeping wheel scroll — useful if accidental clicks interfere with your terminal workflow.
  • Voice dictation fixes: macOS no longer captures silence in long sessions after the default input device changes; auto-submit now fires for languages written without spaces (Japanese, Chinese, Thai); and Linux voice mode now distinguishes "no microphone" from "SoX not installed" when SoX is present but no capture device exists.
  • Tighter external-plugin install consent: external plugins enabled only by a project's .claude/settings.json now require explicit install consent on every loader path — closing a gap where a project config could enable a plugin without prompting. /plugin Enable/Disable also now works when a plugin's plugin.json name differs from its marketplace entry name.
  • Background-agent reliability: background jobs no longer disappear from claude agents (or lose data) when written by a newer Claude Code version; reopening a crashed background task no longer shows a blank screen for up to 5 seconds; and background-agent daemons no longer run unreachable — and block restarts — when the control socket fails to start.
  • Smaller niceties: the claude agents completed list fills available vertical space (the header compacts on short terminals so live sessions stay visible), and Remote session startup now shows a provisioning checklist while the container boots.

Release notes → Affects: /openclaw/, /openclaw/configuration/, /openclaw/skills-guide/, /openclaw/security/, /claude-cowork/, /claude-cowork/setup/

2026-06-27 NemoClaw main

A busy day on main, mostly E2E/CI test stabilization, but with a handful of changes readers running NemoClaw will actually feel. No new version tag landed (NemoClaw stays at v0.0.67):

  • Deep Agents CLI aliases (#5881): a new nemo-deepagents bin runs the LangChain Deep Agents Code runtime directly, and onboarding now accepts short --agent aliases — dcode, deepagents, deepagents-code, langchain — instead of the full langchain-deepagents-code name. (Follow-up #5883 tightened alias parity coverage.)
  • destroy now wipes per-sandbox persistent state (#5449/#5455): nemoclaw <name> destroy used to delete the pod but leave its persistent volume intact, so re-onboarding with the same name resurrected old workspace files (USER.md, SOUL.md, …). Destroy now wipes that durable state, restoring the documented "clean workspace on re-onboard" contract.
  • Host-authoritative shields status (#5838): the in-sandbox /nemoclaw shields command no longer reports UP/DOWN/NOT CONFIGURED from unrelated plugin-local state. Status now points to the authoritative host CLI (nemoclaw <name> shields status), so you can't be misled by a posture only the host can verify.
  • Ollama non-interactive fix (#5898): the Ollama tools-capability gate had a hidden dependency on the onboarding entrypoint that broke --non-interactive and --yes; onboarding now supplies its interaction state explicitly, so scripted/headless Ollama onboarding behaves correctly.
  • Docs: the network-request approval walkthrough now documents the OpenShell TUI sandbox-selection flow and the a/x/A Network Rules keybindings (#5886), and the raw OpenShell policy round-trip examples were aligned to the pinned OpenShell 0.0.44 command surface — policy get <name> --full export, metadata-header strip, name-first policy set (#5880).

Commit log → Affects: /nemoclaw/, /nemoclaw/setup/, /nemoclaw/policy/, /nemoclaw/skills/, /nemoclaw/switching-providers/

See all releases

Browse the full changelog index for the complete history across all platforms, or the daily one-liner for the most recent state of each agent.